Computrons per watt

Mar 12, 2012 8 Replies

(Posted to s.e.basics by mistake.)



A client of mine is trying to build head mounted displays for the DARPA SCENICC program.



formatting link



It is of course insanely ambitious--eye-resolution-limited imaging over the full visual field at 60 fps or so. It also has insanely ambitious SWaP (size, weight, and power) goals: 250 grams including battery, eyeglass or goggle form, power dissipation < 1W.



To try to make a sizing for that power dissipation, could you digital folks give me an idea of how much computing power I can get for, say, half a watt? I'm okay with FPGAs, ASICs, ARMs, GPUs, you name it, but it has to be able to render about 10 Mpel at 60 frames/s. Lots of stuff can be precomputed, but not everything, obviously.



Thanks



Phil Hobbs


Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net

formatting link

These folks have just introduced an M4 version.

f

I believe NVIDIA claim their Tegra 2 can do 1080p at

Thanks. I'm pretty sure this isn't in the ARM Cortex class of problem, but I'd love to be proven wrong.

Cheers

Phil Hobbs

Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net

Your best bet is probably a SoC that goes into mobile phones and tablets. Freescale, Samsung and TI make such SoCs.

Failure does not prove something is impossible, failure simply indicates you are not using the right tools... nico@nctdevpuntnl (punt=.) --------------------------------------------------------------

How much are you going to _do_ to this stream of video that's going by at

300Mpel/second? That's a whole lot of computation, and is going to go beyond any little processors.

I'd go digging on Xilinx and Altera sites, looking for papers on reducing power consumption. Whetever else you do, remember three things: (a), this looks to your humble correspondent like a problem for FPGAs, (b), FPGAs are typically power-hungry, and (c), (b) is strongly affected by the cleverness and dedication of your design team.

Good luck. You're going to have more trouble getting the pixels into and out of your computational device than you will actually manipulating them.

My liberal friends think I'm a conservative kook. My conservative friends think I'm a liberal kook. Why am I not happy that they have found common ground? Tim Wescott, Communications, Control, Circuits & Software http://www.wescottdesign.com

=20

Just say no.

The required refresh takes many watts. Like about 50 to 100. That is a modern display card pushed really hard.

Maybe a cable to a much bigger and more POWERful box.

Other than that is sounds like fun.

?-)

If you fancy FPGA route you are better off with Lattice for low power - since they bought Silicon Blue they are way ahead of Altera or Xilinx on ultra low power. You will need to do a lot of work to get to an accurate estimation of FPGA power for your application.

Michael Kellett

I'd say yes. They want viable plans. They give money to try the most promising plans. There is no need to deliver an actual product. Its about getting paid to do research.

Failure does not prove something is impossible, failure simply indicates you are not using the right tools... nico@nctdevpuntnl (punt=.) --------------------------------------------------------------

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required